DiscussionThe method described in t

Discussion
The method described in this paper gives an aggregate measure of operator “performance”. Results are strongly influenced by dominant taxa in the sample, however, water quality indices based on weighted average equations are also biased towards dominant taxa. Correct identification of all taxa present in a sample is clearly desirable and it is possible that mis-identifications of rare taxa will not be detected using this method. However, over the course of a monitoring programme, the dominant taxa are likely to vary considerably so operators will be tested in a wider range of taxa than may be abundant in a single sample. It is also important that the audit procedure includes a subjective scrutiny of the results and disparities between the two counts are highlighted.
Diatom slides, even when prepared with care, show considerable variation in abundance of taxa (Kelly, 1997) and further work will be necessary to address aspects of this variability. Of particular importance are factors such as non-separation of valves or colonies which leads to contagious distributions and which complicates the use of parametric statistics. However, even if such problems were to be resolved satisfactorily, one would still be faced with the statistical problem of having to make a series of comparisons between proportions of taxa found in the two counts with associated risks of Type II error.
The use of a finite number of well-defined taxa simplifies the process of auditing results. Associated issues include the use of recommended floras and systems of nomenclature. That taxonomic confusion exists within some of these categories is appreciated (Kelly, 1999) and efforts to refine the TDI by increasing the number of taxa identified must accept that greater effort will be required to harmonise taxonomy.
The relationship between similarity and species diversity (Fig. 3) introduces a further variable that needs to be considered when assessing operator performance not just for diatoms but for other taxonomic groups as well. On the basis of data presented here, a Bray–Curtis value of 60% should be acceptable for a sample with a high value of Hill's N2 diversity, but not for one with very low diversity (e.g. where a single taxon dominates the sample), where a higher Bray–Curtis similarity value should be expected. For practical purposes, a threshold Hill's N2 diversity value of 3 is used for the TDI, above which a Bray–Curtis value of 70% is acceptable.
An audit procedure such as this has a number of possible applications within regulatory organisations as it provides an objective measure of operator performance that can be incorporated into QA procedures wherever quantitative, community-based data are to be compared. The Bray–Curtis measure has been used as part of QA procedures encompassing benthic estuarine invertebrates (Unicomarine Ltd, 1996), river phytoplankton and benthic diatoms (Kelly, 1999). In each case, the similarity measure is used alongside subjective comments and notes communicated directly to the analyst in question, so that the exercise is one of ongoing training and development, rather than criticism or bureaucratic scrutiny. The audit is but one part of an overall quality management system, including the use of standard operating procedures and internal quality control measures, designed to reduce process variance (sensuCairns and Smith, 1994) and thus increase the sensitivity of the method to environmental perturbations.